Monthly US Employment Analysis - March 2020

March 2020 Monthly US Employment Analysis

Edward Leamer, Professor Emeritus, UCLA Anderson School
April 3, 2020

This monthly commentary on the payroll data has been based on the premise that to interpret the current data we need to combine it with the recent past, thus creating a smoother, more reliable indicator of where we are. This month’s payroll release of a decline in payroll jobs by 701,000 is an 11-sigma event compared with the recent past — totally different from the February data — and there is no need to look at the recent past to interpret the recent data. We are in trouble here.

It’s worse than these data suggest. These payroll data refer to the payroll period through March 12. The data, thus, account for the lack of hiring that might have preceded March 12, but do not encompass the terminations from March 13 onward. The huge increase in weekly initial claims for unemployment insurance occurred in the weeks after March 12. That’s why we are seeing a loss “only” of 701,000 jobs, not a number close to the 10 million suggested by the claims for unemployment. The worst is yet to come.

Figure 1

The figure below illustrates the annualized percent change in total payrolls since 1970, with the official recession periods shaded. The March 2020 number of -5.5% (annualized) has never occurred during expansions, but there are 12 months during the recessions in which the decline was greater. Except for the 2008–09 recession, these extreme recession dips occurred long ago when the U.S. economy was more volatile, during the two recessions in the 1970s and also in May 1980. That is a warning about just how serious this situation is for the economy, even with the flawed data.

Figure 2

Sectoral detail

Most recessions have very serious job loss in construction and manufacturing, but this shutdown is focused on services that are delivered person to person, with a snowballing soon to come that will affect construction and manufacturing and everything else.

The table below has the data for the five goods sectors and the seven service sectors, with the totals in the bottom row. The first column has last month’s data, and the next three columns have the current release in terms of the change in payrolls, the percent change and the annualize percent change. The total declined by 701,000 jobs, which is a .46% decline of a 5.5% annualized decline.

It’s important to get clear whether the data have been annualized. Quarterly GDP data are reported at annualized rates, which means that the actual GDP in any quarter is one-fourth of the reported numbers. Quarterly growth of GDP is usually normalized, currently around 2%. If you hear about a GDP forecast for the second quarter of minus 20%, that is most likely an annualized number, which means that the actual decline is only about 5%. You can see that reflected in the last row, which has a -0.46% decline in actual payrolls, a -5.5% annualized decline.

The last three columns offer a statistical calculation that reflects how unusual these data are. The mean and standard deviation of the data two years after the 2008–09 recession are in two columns. The last column is the Z-stat, which is the difference between the March 2020 percent change and the historical percent change, divided by the standard deviation. A 2-sigma or more event occurs about 5% of the time. The total payroll decline, when measured against the data from the most recent expansion, is an 11-sigma event.

The Z-stats are color-coded red, yellow and green, from worst to best. Not surprisingly, the big news here is in leisure and hospitality, which had an 18-sigma decline. Next come education and health services (-5.7 sigma) and professional and business services (-4.13). The problems have already leaked into construction and nondurable manufacturing. Government jobs grew more rapidly than their historical mean. Otherwise, only information services had a positive Z-stat, though only slightly greater than the mean. The other happy green shades apply to the relatively small negative Z-stats in mining and logging, and in durable manufacturing. Notice that mining and logging had the second largest percentage decline, but show by far the biggest standard error, so this decline is only a 1-sigma event.

Stay tuned for more bad news.

Figure 3

UCLA Anderson Forecast